oomd.conf

systemd

oomd.conf

oomd.conf.d

Global

systemd-oomd

configuration files

/etc/systemd/oomd.conf, /run/systemd/oomd.conf, /usr/local/lib/systemd/oomd.conf, /usr/lib/systemd/oomd.conf, /etc/systemd/oomd.conf.d/*.conf, /run/systemd/oomd.conf.d/*.conf, /usr/local/lib/systemd/oomd.conf.d/*.conf, /usr/lib/systemd/oomd.conf.d/*.conf

Description

These files configure the various parameters of the systemd(1) userspace out-of-memory (OOM) killer, systemd-oomd.service(8). See systemd.syntax(7) for a general description of the syntax.

Prekill event

systemd-oomd supports notifying external components before killing a control group. This is done by sending a notification over varlink to all sockets found in /run/systemd/oomd.prekill.hook/ folder. Each socket should implement the io.systemd.oom.Prekill interface. The notification contains the control group path to allow the hook to identify which control group is being killed. This allows external components to perform any necessary cleanup or logging before the control group is terminated. The hook is not intended as a way to avoid the kill, but rather as a notification mechanism. Note that this is a privileged option as, even if it has a timeout, is synchronous and delays the kill, so use with care. The typically preferable mechanism to process memory pressure is to do what Resource Pressure Handling describes which is unprivileged, asynchronous and does not delay the kill.

OOM Rulesets

systemd-oomd supports custom rulesets that define conditions and actions for OOM handling on a per-unit basis. Ruleset files use the .oomrule extension and are loaded from /etc/systemd/oomd/rules.d/, /run/systemd/oomd/rules.d/, /usr/local/lib/systemd/oomd/rules.d/, and /usr/lib/systemd/oomd/rules.d/. Units opt into rulesets via the OOMRules= setting in systemd.resource-control(5), which takes a space-separated list of ruleset names (the file name without the .oomrule extension).

Each ruleset file contains a [Rule] section with the following options. At least one of MemoryPressureAbove= or SwapUsageMax= must be configured; rulesets with no conditions are ignored. If both are set, the conditions are combined with AND, i.e. the action is only triggered when both thresholds are exceeded simultaneously.

MemoryPressureAbove=
Sets the memory pressure threshold above which the rule's action will be triggered. The memory pressure represents the fraction of time in a 10 second window in which all tasks in the control group were delayed (PSI full avg10). Takes a value specified in percent (when suffixed with %), permille (‰) or permyriad (‱), between 0% and 100%, inclusive. If unset, this condition is not evaluated. A value of 100% can never be exceeded and is therefore rejected with a warning; a value of 0% makes the condition true on any observed pressure, which is usually not useful.

SwapUsageMax=
Sets the system-wide swap usage threshold above which the rule's action will be triggered. Takes a value specified in percent (when suffixed with %), permille (‰) or permyriad (‱), between 0% and 100%, inclusive. If unset, this condition is not evaluated. A value of 100% can never be exceeded and is therefore rejected with a warning; a value of 0% fires as soon as any swap is in use, which is usually not useful.

Action=
Specifies the action to take when the rule's conditions are met. Takes one of kill-all, kill-by-pgscan, or kill-by-swap. This setting is mandatory; rulesets without Action= are ignored.

kill-all sends SIGKILL to every process in the unit's cgroup hierarchy, including any descendant cgroups.
kill-by-pgscan selects and kills the descendant cgroup with the highest recent page scan (reclaim) rate.
kill-by-swap selects and kills the descendant cgroup with the highest swap usage.

LastingSec=
Sets the duration the conditions must be continuously met before the action is taken. Takes a time span value, see systemd.time(7) for details on the permitted syntax. Defaults to 0, i.e. the action is taken immediately when the conditions are met.

[OOM] Section Options

The following options are available in the [OOM] section:

SwapUsedLimit=
Sets the limit for memory and swap usage on the system before systemd-oomd will take action. If the fraction of memory used and the fraction of swap used on the system are both more than what is defined here, systemd-oomd will act on eligible descendant control groups with swap usage greater than 5% of total swap, starting from the ones with the highest swap usage. Which control groups are monitored and what action gets taken depends on what the unit has configured for ManagedOOMSwap=. Takes a value specified in percent (when suffixed with "%"), permille ("‰") or permyriad ("‱"), between 0% and 100%, inclusive. Defaults to 90%.

DefaultMemoryPressureLimit=
Sets the limit for memory pressure on the unit's control group before systemd-oomd will take action. A unit can override this value with ManagedOOMMemoryPressureLimit=. The memory pressure for this property represents the fraction of time in a 10 second window in which all tasks in the control group were delayed. For each monitored control group, if the memory pressure on that control group exceeds the limit set for longer than the duration set by DefaultMemoryPressureDurationSec=, systemd-oomd will act on eligible descendant control groups, starting from the ones with the most reclaim activity to the least reclaim activity. Which control groups are monitored and what action gets taken depends on what the unit has configured for ManagedOOMMemoryPressure=. Takes a fraction specified in the same way as SwapUsedLimit= above. Defaults to 60%.

DefaultMemoryPressureDurationSec=
Sets the amount of time a unit's control group needs to have exceeded memory pressure limits before systemd-oomd will take action. A unit can override this value with ManagedOOMMemoryPressureDurationSec=. Memory pressure limits are defined by DefaultMemoryPressureLimit= and ManagedOOMMemoryPressureLimit=. Must be set to 0, or at least 1 second. Defaults to 30 seconds when unset or 0.

PrekillHookTimeoutSec=
Sets the amount of time systemd-oomd will wait for pre-kill hooks to complete, before proceeding with the control group termination. Pre-kill hooks work by placing varlink socket to /run/systemd/oomd.prekill.hook/ folder. Each socket should implement interface for notification to work. systemd-oomd sends a notification before killing a control group for each discovered socket. The timeout is intended to be global and not per hook. If all hooks return earlier, the kill is performed as soon as possible. The timeout must be at least 1s. Defaults to 0, which means systemd-oomd will not wait and no notifications will be sent.